A Two-Stage, Object-Centric Deep Learning Framework for Robust Exam Cheating Detection

2026-04-21 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, extended

Summary

A new two-stage, object-centric deep learning framework has been developed for robust exam cheating detection, addressing the inefficiencies and limitations of human invigilation and existing AI systems. The framework utilizes a YOLOv8n model to localize students in exam-room images, followed by a fine-tuned RexNet-150 model to classify cropped regions as either normal or cheating behavior. Trained on a dataset compiled from 10 independent sources with 273,897 samples, the system achieved 0.95 accuracy, 0.94 recall, 0.96 precision, and 0.95 F1-score, representing a 13% increase over a baseline accuracy of 0.82 in video-based detection. With an average inference time of 13.9 ms per sample, the approach is scalable for large-scale deployment and incorporates ethical considerations by delivering private outcomes to students.

Key takeaway

For research scientists developing AI-powered proctoring systems, you should consider adopting a two-stage, object-centric framework to enhance detection accuracy and scalability. Focusing computational analysis on individual examinees, rather than full-frame analysis, demonstrably improves performance by reducing background noise. Prioritize curating diverse, large-scale datasets to build more generalizable and robust models, and explore multi-class classification for more granular insights into cheating behaviors.

Key insights

A two-stage, object-centric deep learning framework significantly improves exam cheating detection accuracy and scalability.

Principles

• Isolate subjects to reduce background noise.
• Decouple detection and classification tasks.
• Curate large, standardized datasets for robustness.

Method

The method involves YOLOv8n for student localization, followed by cropping and preprocessing regions of interest, then classifying behavior using a fine-tuned RexNet-150 model.

In practice

• Use YOLOv8n for efficient object detection.
• Employ RexNet-150 for robust behavior classification.
• Implement early-stopping based on validation F1-score.

Topics

• Exam Cheating Detection
• Deep Learning Frameworks
• YOLOv8n
• RexNet-150
• Object-Centric Analysis

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• A Two-Stage, Object-Centric Deep Learning Framework for Robust Exam Cheating Detection — A two-stage deep learning framework improves exam cheating detection with high accuracy and ethical private feedback.
• Teacher-Student Structure for Domain Adaptation in Ensemble Audio-Visual Video Deepfake Detection — A teacher-student framework enhances ensemble audio-visual deepfake detection models for robust generalization across diverse, unseen domains.
• How Stripe Detects Fraudulent Transactions Within 100 ms — Evolving ML architectures and leveraging vast network data are crucial for high-performance fraud detection.
• An Adaptive Data cleaning Framework for Noisy Label Detection — A self-adaptive framework integrates multi-metric cues for robust noisy-label detection, avoiding manual thresholds and noise priors.
• Finding Bad Image Data using UMAP and Prodigy — UMAP-driven dimensionality reduction combined with human-in-the-loop annotation effectively identifies bad examples in large image datasets.
• Federated orthogonal learning for detection of liver lesions from multi-phase contrast-enhanced CT images — FedOG enhances federated learning for medical image segmentation by using orthogonal gradient decomposition to counter data heterogeneity and incomplete phases.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.